> My 30k ft view is that the stock will inevitably slide as AI datacenter spending goes down.

Their stock trajectory started with one boom (cryptocurrencies) and then seamlessly progressed to another (AI). You're basically looking at a decade of "number goes up". So yeah, it will probably come down eventually (or the inflation will catch up), but it's a poor argument for betting against them right now.

Meanwhile, the investors who were "wrong" anticipating a cryptocurrency revolution and who bought NVDA have not much to complain about today.

▲

mysteria 12 hours ago | parent | next [-]

Personally I wonder even if the LLM hype dies down we'll get a new boom in terms of AI for robotics and the "digital twin" technology Nvidia has been hyping up to train them. That's going to need GPUs for both the ML component as well as 3D visualization. Robots haven't yet had their SD 1.1 or GPT-3 moment and we're still in the early days of Pythia, GPT-J, AI Dungeon, etc. in LLM speak.

▲

iwontberude 10 hours ago | parent [-]

Exactly, they will pivot back to AR/VR

▲

mysteria 10 hours ago | parent [-]

That's going to tank the stock price though as that's a much smaller market than AI, though it's not going to kill the company. Hence why I'm talking about something like robotics which has a lot of opportunity to grow and make use of all those chips and datacenters they're building.

Now there's one thing with AR/VR that might need this kind of infrastructure though and that's basically AI driven games or Holodeck like stuff. Basically have the frames be generated rather than modeled and rendered traditionally.

	▲	bigyabai 10 hours ago \| parent [-]
		Nvidia's not your average bear, they can walk and chew bubblegum at the same time. CUDA was developed off money made from GeForce products, and now RTX products are being subsidized by the money made on CUDA compute. If an enormous demand for efficient raster compute arises, Nvidia doesn't have to pivot much further than increasing their GPU supply. Robotics is a bit of a "flying car" application that gets people to think outside the box. Right now, both Russia and Ukraine are using Nvidia hardware in drones and cruise missiles and C2 as well. The United States will join them if a peer conflict breaks out, and if push comes to shove then Europe will too. This is the kind of volatility that crazy people love to go long on.

▲

munk-a 12 hours ago | parent | prev | next [-]

That's the rub - it's clearly overvalued and will readjust... the question is when. If you can figure out when precisely then you've won the lottery, for everyone else it's a game of chicken where for "a while" money that you put into it will have a good return. Everyone would love if that lasted forever so there is a strong momentum preventing that market correction.

▲

jama211 12 hours ago | parent [-]

It was overvalued when crypto was happening too, but another boom took its place. Of course, lightening rarely strikes twice and all that, but it proves overvalued doesn’t mean the price is guaranteed to go down it seems. Predicting the future is hard.

▲

pixl97 9 hours ago | parent | next [-]

As they say, the market can remain irrational far longer than you can remain solvent.

	▲	jama211 2 hours ago \| parent [-]
		Hah! Indeed

▲

sidrag22 11 hours ago | parent | prev [-]

if there was anything i was going to bet against between 2019 and now, it was nvidia... and wow it feels wild how much in the opposite direction it went.

I do wonder what people would think the reasoning would be for them to increase in value this much back then, prolly would just assume crypto related still.

	▲	jama211 2 hours ago \| parent [-]
		It’s not impossible they could’ve seen AI investment coming but it would’ve been very hard

▲

ericmcer 12 hours ago | parent | prev [-]

Crypto & AI can both be linked to part of a broader trend though, that we need processors capable of running compute on massive sets of data quickly. I don't think that will ever go down, whether some new tech emerges or we just continue shoveling LLMs into everything. Imagine the compute needed to allow every person on earth to run a couple million tokens through a model like Anthropic Opus every day.

	▲	pixl97 9 hours ago \| parent [-]
		Agreed, single thread performance increases are dead and things are moving to massively parallel processing.